As datasets becomes bigger, more samples, more msruns we should move out of xml, csv txt files, and more into hdf5, parquet,sqlite etc.